本文章由公号【开发小鸽】发布!欢迎关注!!!
老规矩–妹妹镇楼:
一. 自定义Harris角点检测器
- 基于Harris角点检测
- 首先通过计算矩阵M得到两个特征值,然后计算得到角点响应值
- 设置阈值,通过阈值计算得到有效响应的角点位置
二. API分析
cv::cornerEigenValsAndVecs
自定义角点检测器
cv::cornerEigenValsAndVecs(
InputArray src,
OutputArray dst,
int blockSize,
int kSize,
int borderType=BORDER_DEFAULT
)
二. 代码实现:
/*****自定义Harris角点检测器*****/
#include<opencv2/opencv.hpp>
#include<iostream>
#include<math.h>
using namespace std;
cv::Mat src, gray_src;
//最小Harris响应
double harris_min_rsp;
double harris_max_rsp;
int qualityLevel = 30;
int max_count = 100;
//Harris结果
cv::Mat harris_dst;
//每个像素响应的值存储
cv::Mat harrisRspImg;
void CumstomHarris_Demo(int, void*);
int main(int argc, char** argv) {
src = cv::imread("建筑.jpg");
cv::namedWindow("in", cv::WINDOW_NORMAL);
cv::imshow("in", src);
cv::cvtColor(src, gray_src, cv::COLOR_BGR2GRAY);
//计算特征值
int blockSize = 3;
int kSize = 3;
double k = 0.04;
//3个通道,每个通道两个值
harris_dst = cv::Mat::zeros(src.size(), CV_32FC(6));
//角点计算
cv::cornerEigenValsAndVecs(gray_src, harris_dst, blockSize, kSize, 4);
//计算响应
harrisRspImg = cv::Mat::zeros(src.size(), CV_32FC1);
for (int row = 0; row < harris_dst.rows; ++row) {
for (int col = 0; col < harris_dst.cols; ++col) {
double lambda1 = harris_dst.at<cv::Vec6f>(row, col)[0];
double lambda2 = harris_dst.at<cv::Vec6f>(row, col)[1];
//公式计算响应值
harrisRspImg.at<float>(row, col) = lambda1 * lambda2 - k * pow((lambda1 + lambda2), 2);
}
}
//查找最小值,最大值的位置
cv::minMaxLoc(harrisRspImg, &harris_min_rsp, &harris_max_rsp, 0, 0, cv::Mat());
cv::namedWindow("ou", cv::WINDOW_NORMAL);
cv::createTrackbar("Quality Value:", "ou", &qualityLevel, max_count,
CumstomHarris_Demo);
CumstomHarris_Demo(0, 0);
cv::waitKey(0);
return 0;
}
void CumstomHarris_Demo(int, void*) {
if (qualityLevel < 10)
qualityLevel = 10;
cv::Mat resultImg = src.clone();
//响应值的阈值
float t = harris_min_rsp + (((double)qualityLevel) / max_count) *
(harris_max_rsp - harris_min_rsp);
for (int row = 0; row < src.rows; ++row) {
for (int col = 0; col < src.cols; ++col) {
//响应值只有大于阈值才是角点
float v = harrisRspImg.at<float>(row, col);
if (v > t) {
//画出角点,关键,col,row是反过来的,因为图像的横坐标是图像的列,纵坐标是图像的行
circle(resultImg, cv::Point(col, row), 2, cv::Scalar(0, 0, 255),
2, 8, 0);
}
}
}
cv::imshow("ou", resultImg);
}